Hot Take Some syncable passkeys are hard to comprehend and may be untrustworthy (Pass the Passkey: A Novel Attack Surface in Passwordless Authentication)

Wrecker4923

Level 9
Thread author
Verified
Well-known
Apr 11, 2024
413
2,232
769

Original Report:​

Summary Report from Bleeping Computer​

My Takes​

  • I love the idea of device-bound passkeys, but I’m wary about syncable passkeys—especially for important accounts, as passkey authentication usually doesn’t require an additional authentication.
  • Regardless, I think typical passkey registration and authentication are still straightforward, which makes it easy for some users to feel confident about why it’s safer and easier than normal passwords plus phishable 2FA.
  • Google password manager introduces a cloud-based authenticator (syncable), which makes registration and authentication harder to understand and exposes more attack surface, just like this article shows. At this point, I’m not even trying to understand it.
  • As far as I know, device-bound passkeys (Security keys, Windows Hello) are still unaffected by the issues mentioned in this report.
  • My take is to use device-bound passkeys as much as possible, and use syncable passkeys—especially in Google Password Manager—as little as possible.

One-Paragraph Summary​

Palo Alto Networks Unit 42's report, "Pass the Passkey: A Novel Attack Surface in Passwordless Authentication," exposes how malware running on a compromised endpoint can exploit onboarding, recovery, and device-trust workflows in Google Cloud Authenticator to execute full account takeovers of passkey-protected accounts. By demonstrating three distinct attack vectors collectively named "Pass-ta-key", the researchers show that unprivileged endpoint malware can bypass user consent, user verification (such as biometrics/PINs), and hardware-backed TPM protections to forge assertions, trick cloud authenticators, or exfiltrate plaintext passkey private keys. These findings reveal that while passkeys effectively eliminate traditional password phishing and credential-dumping attacks, they shift the threat model toward endpoint security and cloud authenticator implementation risks.

Key Takeaways & Important Details​

1. Stage Zero Reconnaissance (Local Data Harvesting)​

  • Unprivileged Access: Chrome stores synced passkey records locally within %LocalAppData%\Google\Chrome\User Data\<Profile>\Sync Data\LevelDB as proto-encoded WebauthnCredentialSpecifics records.
  • Target Identification: Without requiring elevated administrator privileges, local malware can enumerate targeted accounts, usernames, credential IDs, and encrypted private key structures before launching an attack.

2. The Three "Pass-ta-key" Attack Classes​

  • Pass-ta-key Attack (Device Identity Impersonation):
    • Mechanism: Chrome stores its TPM-wrapped identity private key (wrapped_identity_private_key) inside passkey_enclave_state. Malware reads this blob and uses standard Windows Cryptography API (CNG) calls (NCryptSignHash) to issue signing requests to the TPM without triggering device-unlock prompts or administrative elevation.
    • Impact: The malware initiates a WebSocket session with Google Cloud Authenticator, obtains valid signed assertions from the cloud enclave, and logs in as the victim without biometrics, PINs, or user interaction.
  • Silver Pass-ta-key Attack (Bypassing User Verification / Biometrics):
    • Mechanism: Exploits weaknesses in how device trust and user verification (UV) state are asserted to the cloud authenticator.
    • Impact: The attacker deceives the cloud authenticator into believing that biometric or PIN verification succeeded on the client. Once the initial token is obtained, the attacker can complete the account takeover remotely from their own machine without needing ongoing access to the victim's physical device.
  • Golden Pass-ta-key Attack (Exfiltrating Synced Private Keys):
    • Mechanism: Misuses device onboarding and key recovery workflows within the cloud authenticator architecture.
    • Impact: Allows the attacker to extract raw, synced passkey private keys in plaintext. Once exfiltrated, these keys can be copied, exported, or traded on cybercrime marketplaces, completely undermining the "non-transferable" premise of passkey credentials.

3. Defensive & Security Implications​

  • Shift in the Threat Surface: Passkeys eliminate server-side database leaks and phishing, but they move the battleground to client endpoints and identity sync architectures.
  • TPM & Software Limits: Hardware-backed protections (like TPMs) only protect keys from raw extraction if software API layer restrictions or user-presence enforcement are improperly designed.
  • Mitigation Focus: Security teams must prioritize endpoint detection and response (EDR/MDR), identity security controls, and strict privilege management to detect local key usage anomalies before malware can hijack cloud authenticator workflows.
 
maybe a USB key is an improvement I'm not sure, but both I reckon better than passwords even if 2FA with a phone is used????
As far as I can tell, Windows Hello implementation is pretty solid. If it’s implemented the way I understand it, it’s practically as safe as a security key (with a larger exposure to side-channel attacks). The problem is that your TPM can become unsynced from your Windows state (from bad corruption, restorations, updates, etc.), and if that happens, it will wipe all the passkeys stored on the computer. Security keys don’t have this problem, but they’re expensive, and they can still fail.

To authenticate using Windows Hello, you need the computer "with the TPM" intact, and you need to either provide biometrics or a PIN. That’s already 2FA. It’s unphishable as long as your Windows and browser components are intact; you can’t override it, unlike a password manager.
 
I myself hate passkeys, they are insecure by design, but they are not optional, they are forced and they can not be disabled. I would rather login using 2FA every single time!

capture_08082026_224417.jpg
 
An extreme case scenario? I'm still okay with Proton Pass and 2FAS.
Most of my logins still don’t have passkeys. I’m still using a TOTP authenticator, too.

People use passkeys either for their security (or perceived security😅) or for convenience. For example, on MT, whenever I got logged out unexpectedly—or used to have to supply the TOTP code monthly—I now use a fingerprint reader to authenticate. That’s convenient; it’s hard to go back.

For important accounts (like Google, etc.), I'd love to delete the TOTP authentication altogether if other 2FAs are more predictable. Besides being phishable, TOTP 2FA is brute-forceable if not "perfectly" rate-limited, as demonstrated by Dashlane's vault breaches recently.